A machine learning approach to predict daptomycin exposure from two concentrations based on Monte Carlo simulations.

Daptomycin is a concentration-dependent lipopeptide antibiotic for which exposure/effect relationships have been shown. Machine learning (ML) algorithms, developed to predict the individual exposure to drugs, have shown very good performances in comparison to maximum a posteriori Bayesian estimation (MAP-BE). The aim of this work was to predict the area under the blood concentration curve (AUC) of daptomycin from two samples and a few covariates using XGBoost ML algorithm trained on Monte Carlo simulations. Five thousand one hundred fifty patients were simulated from two literature population pharmacokinetics models. Data from the first model were split into a training set (75%) and a testing set (25%). Four ML algorithms were built to learn AUC based on daptomycin blood concentration samples at pre-dose and 1 h post-dose. The XGBoost model (best ML algorithm) with the lowest root mean square error (RMSE) in a 10-fold cross-validation experiment was evaluated in both the test set and the simulations from the second population pharmacokinetic model (validation). The ML model based on the two concentrations, the differences between these concentrations, and five other covariates (sex, weight, daptomycin dose, creatinine clearance, and body temperature) yielded very good AUC estimation in the test (relative bias/RMSE = 0.43/7.69%) and validation sets (relative bias/RMSE = 4.61/6.63%). The XGBoost ML model developed allowed accurate estimation of daptomycin AUC using C0, C1h, and a few covariates and could be used for exposure estimation and dose adjustment. This ML approach can facilitate the conduct of future therapeutic drug monitoring (TDM) studies.

Optimization of Ganciclovir and Valganciclovir Starting Dose in Children by Machine Learning.

BACKGROUND AND OBJECTIVES: Ganciclovir (GCV) and valganciclovir (VGCV) show large interindividual pharmacokinetic variability, particularly in children. The objectives of this study were (1) to develop machine learning (ML) algorithms trained on simulated pharmacokinetics profiles obtained by Monte Carlo simulations to estimate the best ganciclovir or valganciclovir starting dose in children and (2) to compare its performances on real-world profiles to previously published equation derived from literature population pharmacokinetic (POPPK) models achieving about 20% of profiles within the target. MATERIALS AND METHODS: The pharmacokinetic parameters of four literature POPPK models in addition to the World Health Organization (WHO) growth curve for children were used in the mrgsolve R package to simulate 10,800 pharmacokinetic profiles. ML algorithms were developed and benchmarked to predict the probability to reach the steady-state, area-under-the-curve target (AUC0-24 within 40-60 mg × h/L) based on demographic characteristics only. The best ML algorithm was then used to calculate the starting dose maximizing the target attainment. Performances were evaluated for ML and literature formula in a test set and in an external set of 32 and 31 actual patients (GCV and VGCV, respectively). RESULTS: A combination of Xgboost, neural network, and random forest algorithms yielded the best performances and highest target attainment in the test set (36.8% for GCV and 35.3% for the VGCV). In actual patients, the best GCV ML starting dose yielded the highest target attainment rate (25.8%) and performed equally for VGCV with the Franck model formula (35.3% for both). CONCLUSION: The ML algorithms exhibit good performances in comparison with previously validated models and should be evaluated prospectively.

Better Rejection-Free Survival at Three Years in Kidney Transplant Recipients With Model-Informed Precision Dosing of Mycophenolate Mofetil.

The clinical impact of individual dose adjustment of mycophenolate mofetil is still debated, due to conflicting results from randomized clinical trials. This retrospective study aimed to compare 3-year rejection-free survival and adverse effects between adult kidney transplant recipients (KTRs) with or without mycophenolate mofetil model-informed precision dosing (MIPD). MIPD is defined here as mycophenolic acid area under the curve (AUC0-12h ) estimation using a limited sampling strategy, pharmacokinetic models and Bayesian estimators; dose recommendation to reach AUC0-12h = 45 mg.h/L; using a widely used online expert system. The study, nested in two multicenter prospective cohort studies, focused on patients who received a mycophenolate drug and were followed up for 1-3 years. Mycophenolate mofetil MIPD was prescribed as per local practice, on a regular basis, when deemed necessary, or not at all. The MIPD group included 341 KTRs and the control group 392. At 3 years, rejection-free survival was respectively 91.2% and 80.6% (P < 0.001) and the cumulative incidence of rejection 5.08% vs. 12.7% per patient × year (hazard ratio = 0.49 (0.34, 0.71), P < 0.001), corresponding to a 2.5-fold reduction. Significant association with rejection-free survival was confirmed in patients at low or high risk of rejection (P = 0.017 and 0.013) and in patients on tacrolimus, but not on cyclosporine (P < 0.001 and 0.205). The mycophenolate mofetil MIPD group had significantly more adverse effects, but most occurred before the first AUC0-12h , suggesting some may be the reason why MIPD was ordered.

Cytomegalovirus detected by qPCR in iris and ciliary body of immunocompetent corneal donors.

BACKGROUND: Cytomegalovirus (CMV) can cause a wide panel of ocular infections. The involvement of CMV as a cause of anterior uveitis in the immunocompetent patient is recent and remains poorly understood. OBJECTIVE: To investigate the presence of CMV in anterior uveal tissues of immunocompetent corneal donors. STUDY DESIGN: We collected aqueous humor, iris, and ciliary body from both eyes of 25 donors died at the Limoges University Hospital between January 2020 and July 2021. CMV serology was determined for all patients from post-mortem blood sample. Ocular tissues were split in 2 fragments for qPCR and 2 for histological analysis. CMV genomes copies were quantified by Multiplex qPCR after DNA extraction. RESULTS: 16 of 25 patients (64%) displayed positive CMV serology, with a median age of 67 years. Viremia was positive in 3 of 16 (19%) CMV-positive patients. No CMV DNA copies were found from the aqueous humor samples. CMV DNA was detected in iris and ciliary body of 28 of 32 eyes of seropositive donors, and 5 of 18 eyes of seronegative donors. The median viral copy number [IQR] was 2.41 × 102 [8.91 × 101 - 1.01 × 103] copies/1 × 106 cells in the CMV-positive group and 0.00 [0.00 - 3.54 × 102] copies/1 × 106 cells in the CMV-negative group (p<0.001). Histology and immunohistochemistry did not reveal any CMV lesions from any sample. CONCLUSION: CMV DNA was found in iris and ciliary body of immunocompetent seropositive patients, but also, although less frequently, from seronegative donors. These results highlight mechanisms of infection, latency and reactivation of CMV in ocular tissues.

[ACACIAS 3: Learning about announcement consultations in the second cycle of medical studies]. / ACACIAS 3 : Apprentissage des consultations d'annonce dans le deuxième cycle d'études médicales.

INTRODUCTION: The second cycle of medical studies is a key time for developing interpersonal skills and the doctor-patient relationship. High-fidelity simulation is an initial learning option that enables learners to confront situations involving empathy. METHODS: This is a feedback report from May 2023 on the implementation of simulation as a training tool for 2nd cycle medical students in the announcement consultation. The training consists of two parts: theoretical teaching via a digital platform with an assessment of theoretical knowledge and a practical part with a simulation session with an actress playing a standardized patient. The acquisition of skills and the reflexivity of learners are assessed by means of a pre- and post-test. RESULTS: Twenty-nine externs took part in this project. Student satisfaction was 96 %. The feedback was very positive, both in terms of the quality of the sessions and the briefings/debriefings. Almost all the students wanted to repeat the experience. The simulation exercise was beneficial for the students in terms of the development (before vs. after) of their skills (verbal, emotional and relational) (1.05±0.25 vs. 1.22±0.19, P=0.047) and appeared to be relevant to the development of reflexivity (3.29±0.72 vs. 3.48±0.9, P=0.134). CONCLUSION: This first published French study demonstrates the feasibility and value of training in announcing a diagnosis, combining teaching via a digital platform and high-fidelity simulation for second cycle medical students.

In Silico Pharmacokinetics Evaluation of Forgiveness for Doravirine and Rilpivirine.

BACKGROUND: This study aimed to evaluate the concentrations of rilpivirine (RLP) and doravirine (DOR) after 3 days-off using simulations from population pharmacokinetics models. METHODS: The authors conducted a series of 500 sets of 10,000 Monte Carlo simulations to examine the steady-state conditions for 2 common dosage levels: 25 mg/d for RLP and 100 mg/d for DOR. These simulations were conducted under 2 scenarios: 1 without drug cessation and another after a 3-day break. The validity of the implementation was established through a comparison of median trough concentrations (C24h) with previously reported data. Subsequently, the proportion of simulated patients with C24h and C72h after 3 days-off (C72h/3do) that exceeded the inhibitory concentration 50 (IC50), 5.2 mcg/L for DOR and 20.5 mcg/L for RLP respectively, was calculated. The inhibitory quotient (IQ) was also computed, which was 6 times IC50 for DOR and 4.5 times IC50 for RLP. Finally, nomograms were constructed to estimate the probability of having C72h/3do > IC50 or > IQ for different ranges of C24h. RESULTS: Simulated C24h median ± SD for RLP were 61.8 ± 0.4 mcg/L and for DOR 397 ± 0 mcg/L. For RLP, 99.3 ± 0.1% exceeded IC50 at C24h, 16.4 ± 0.4% at C72h/3do, and none surpassed the IQ threshold. In contrast, DOR had 100% ± 0% above IC50 at C24h, 93.6 ± 0.2% at C72h/3do, and 58.6 ± 0.5% exceeded the IQ. CONCLUSIONS: These findings suggest that treatment with DOR may offer a more forgiving therapeutic profile than RLP, given the larger proportion of patients achieving effective drug exposure with DOR. However, it is important to acknowledge a significant limitation of this study, namely, the assumption that drug concentration is a perfect surrogate for drug effectiveness.

Tacrolimus population pharmacokinetics in adult heart transplant patients.

INTRODUCTION: Tacrolimus is an immunosuppressant largely used in heart transplantation. However, the calculation of its exposure based on the area under the curve (AUC) requires the use of a population pharmacokinetic (PK) model. The aims of this work were (i) to develop a population PK model for tacrolimus in heart transplant patients, (ii) to derive a maximum a posteriori Bayesian estimator (MAP-BE) based on a limited sampling strategy (LSS) and (iii) to estimate probabilities of target attainment (PTAs) for AUC and trough concentration (C0). MATERIAL AND METHODS: Forty-seven PK profiles (546 concentrations) of 18 heart transplant patients of the Pharmacocinétique des Immunosuppresseurs chez les patients GREffés Cardiaques study receiving tacrolimus (Prograf®) were included. The database was split into a development (80%) and a validation (20%) set. PK parameters were estimated in MONOLIX® and based on this model a Bayesian estimator using an LSS was built. Simulations were performed to calculate the PTA for AUC and C0. RESULTS: The best model to describe the tacrolimus PK was a two-compartment model with a transit absorption and a linear elimination. Only the CYP3A5 covariate was kept in the final model. The derived MAP-BE based on the LSS (0-1-2 h postdose) yielded an AUC bias ± SD = 2.7 ± 10.2% and an imprecision of 9.9% in comparison to the reference AUC calculated using the trapezoidal rule. PTAs based on AUC or C0 allowed new recommendations to be proposed for starting doses (0.11 mg·kg-1 ·12 h-1 for the CYP3A5 nonexpressor and 0.22 mg·kg1 ·12 h-1 for the CYP3A5 expressor). CONCLUSION: The MAP-BE developed should facilitate estimation of tacrolimus AUC in heart transplant patients.

A Hybrid Algorithm Combining Population Pharmacokinetic and Machine Learning for Isavuconazole Exposure Prediction.

OBJECTIVES: Maximum a posteriori Bayesian estimation (MAP-BE) based on a limited sampling strategy and a population pharmacokinetic (POPPK) model is used to estimate individual pharmacokinetic parameters. Recently, we proposed a methodology that combined population pharmacokinetic and machine learning (ML) to decrease the bias and imprecision in individual iohexol clearance prediction. The aim of this study was to confirm the previous results by developing a hybrid algorithm combining POPPK, MAP-BE and ML that accurately predicts isavuconazole clearance. METHODS: A total of 1727 isavuconazole rich PK profiles were simulated using a POPPK model from the literature, and MAP-BE was used to estimate the clearance based on: (i) the full PK profiles (refCL); and (ii) C24h only (C24h-CL). Xgboost was trained to correct the error between refCL and C24h-CL in the training dataset (75%). C24h-CL as well as ML-corrected C24h-CL were evaluated in a testing dataset (25%) and then in a set of PK profiles simulated using another published POPPK model. RESULTS: A strong decrease in mean predictive error (MPE%), imprecision (RMSE%) and the number of profiles outside ± 20% MPE% (n-out20%) was observed with the hybrid algorithm (decreased in MPE% by 95.8% and 85.6%; RMSE% by 69.5% and 69.0%; n-out20% by 97.4% and 100% in the training and testing sets, respectively. In the external validation set, the hybrid algorithm decreased MPE% by 96%, RMSE% by 68% and n-out20% by 100%. CONCLUSION: The hybrid model proposed significantly improved isavuconazole AUC estimation over MAP-BE based on the sole C24h and may improve dose adjustment.

Mycophenolate Mofetil Dose Adjustment in Pediatric Kidney Transplant Recipients.

BACKGROUND: The Immunosuppressant Bayesian Dose Adjustment web site aids clinicians and pharmacologists involved in the care of transplant recipients; it proposes dose adjustments based on the estimated area under the concentration-time curve (AUCs). Three concentrations (T 20 min , T 1 h , and T 3 h ) are sufficient to estimate mycophenolic acid (MPA) AUC 0-12 h in pediatric kidney transplant recipients. This study investigates mycophenolate mofetil (MMF) doses and MPA AUC values in pediatric kidney transplant recipients, and target exposure attainment when the proposed doses were followed, through a large-scale analysis of the data set collated since the inception of the Immunosuppressant Bayesian Dose Adjustment web site. METHODS: In this study, 4051 MMF dose adjustment requests, corresponding to 1051 patients aged 0-18 years, were retrospectively analyzed. AUC calculations were performed in the back office of the Immunosuppressant Bayesian Dose Adjustment using published Bayesian and population pharmacokinetic models. RESULTS: The first AUC request was posted >12 months posttransplantation for 41% of patients. Overall, only 50% had the first MPA AUC 0-12 h within the recommended 30-60 mg.h/L range. When the proposed dose was not followed, the proportion of patients with an AUC in the therapeutic range for MMF with cyclosporine or tacrolimus at the subsequent request was lower (40% and 45%, respectively) than when it was followed (58% and 60%, respectively): P = 0.08 and 0.006, respectively. Furthermore, 3 months posttransplantation, the dispersion of AUC values was often lower at the second visit when the proposed doses were followed, namely, P = 0.03, 0.003, and 0.07 in the 4 months-1 year, and beyond 1 year with <6-month or >6-month periods between both visits, respectively. CONCLUSIONS: Owing to extreme interindividual variability in MPA exposure, MMF dose adjustment is necessary; it is efficient at reducing such variability when based on MPA AUC.

Doravirine Exposure Decreased by Dialysis in a HIV Patient: A Grand Round.

BACKGROUND: The authors report the case of a 66-year-old male patient who was hemodialyzed 3 times per week for chronic renal failure and treated with 100 mg of doravirine once daily in combination with dolutegravir for HIV-1. No dose adjustment is required for doravirine in cases of severe renal injury, but the effect of dialysis on its exposure is poorly understood. METHODS RESULTS: Two series of 2 samples were drawn before and after 4-hour hemodialysis and showed an average doravirine concentration decrease of 48.1 ± 6.7%. The effects of hemodialysis were important, contrary to what was expected and has been previously reported. In addition, intraindividual variability was low. Nevertheless, because the concentrations reported were largely above the inhibitory concentration 50 (IC 50 ), no dose adjustment was required. CONCLUSIONS: The decrease in doravirine concentration due to hemodialysis observed in this case report was quite significant. Therefore, therapeutic drug monitoring might be recommended in certain patients undergoing doravirine treatment also on hemodialysis.

Can the Area Under the Curve/Trough Level Ratio Be Used to Optimize Tacrolimus Individual Dose Adjustment?

BACKGROUND: The aim of this work was to evaluate, in a large data set of renal transplant recipients, the intraindividual variability of the area under the curve (AUC)/predose concentration (C0) ratio in comparison with that of AUC, C0, AUC/dose, and C0/dose. METHODS: Patients with at least 2 tacrolimus AUC estimation requests were extracted from the Immunosuppressant Bayesian dose Adjustment website, and relative variations between 2 consecutive visits for the different metrics were calculated and compared. RESULTS: Data from 1325 patients on tacrolimus (3827 measured C0 and estimated AUC) showed that the lowest mean relative variation between 2 consecutives visits was for the AUC/C0 ratio (95% confidence interval [CI] relative fold change = -43% to 44% for AUC/C0; 95% CI, -77% to 72% for AUC; 95% CI, -82% to 98% for AUC/dose; 95% CI, -81% to 80% for C0 and 95% CI, -94% to 117% for C0/dose. The correlation between 2 consecutive requests, whether close or far apart, was also best for the AUC/C0 ratio ( r = 0.33 and r = 0.34, respectively) in comparison with C0 ( r = 0.21 and r = 0.22, respectively) and AUC ( r = 0.19 and 0.28, respectively). Regression analysis between AUC0-24 and C0 showed that for some patients, the usual C0 targets translated into some very unusual AUC values. As the AUC/C0 ratio is quite stable during large periods, individualized C0 targets can be derived from the AUC targets, and an algorithm that estimates the individualized C0 was developed for situations in which prior AUC estimates are available or not. CONCLUSIONS: In this study, we confirmed in a large data set that the AUC/C0 ratio yields low intraindividual variability, whereas C0 shows the largest, and we propose to calculate individualized C0 targets based on this ratio.

Application of machine learning to predict tacrolimus exposure in liver and kidney transplant patients given the MeltDose formulation.

PURPOSE: Machine Learning (ML) algorithms represent an interesting alternative to maximum a posteriori Bayesian estimators (MAP-BE) for tacrolimus AUC estimation, but it is not known if training an ML model using a lower number of full pharmacokinetic (PK) profiles (= "true" reference AUC) provides better performances than using a larger dataset of less accurate AUC estimates. The objectives of this study were: to develop and benchmark ML algorithms trained using full PK profiles to estimate MeltDose®-tacrolimus individual AUCs using 2 or 3 blood concentrations; and to compare their performance to MAP-BE. METHODS: Data from liver (n = 113) and kidney (n = 97) transplant recipients involved in MeltDose-tacrolimus PK studies were used for the training and evaluation of ML algorithms. "True" AUC0-24 h was calculated for each patient using the trapezoidal rule on the full PK profile. ML algorithms were trained to estimate tacrolimus true AUC using 2 or 3 blood concentrations. Performances were evaluated in 2 external sets of 16 (renal) and 48 (liver) transplant patients. RESULTS: Best estimation performances were obtained with the MARS algorithm and the following limited sampling strategies (LSS): predose (0), 8, and 12 h post-dose (rMPE = - 1.28%, rRMSE = 7.57%), or 0 and 12 h (rMPE = - 1.9%, rRMSE = 10.06%). In the external dataset, the performances of the final ML algorithms based on two samples in kidney (rMPE = - 3.1%, rRMSE = 11.1%) or liver transplant recipients (rMPE = - 3.4%, rRMSE = 9.86%) were as good as or better than those of MAP-BEs based on three time points. CONCLUSION: The MARS ML models developed using "true" MeltDose®-tacrolimus AUCs yielded accurate individual estimations using only two blood concentrations.

Machine learning-supported interpretation of kidney graft elementary lesions in combination with clinical data.

Interpretation of kidney graft biopsies using the Banff classification is still heterogeneous. In this study, extreme gradient boosting classifiers learned from two large training datasets (n = 631 and 304 cases) where the "reference diagnoses" were not strictly defined following the Banff rules but from central reading by expert pathologists and further interpreted consensually by experienced transplant nephrologists, in light of the clinical context. In three external validation datasets (n = 3744, 589, and 360), the classifiers yielded a mean ROC curve AUC (95%CI) of: 0.97 (0.92-1.00), 0.97 (0.96-0.97), and 0.95 (0.93-0.97) for antibody-mediated rejection (ABMR); 0.94 (0.91-0.96), 0.94 (0.92-0.95), and 0.91 (0.88-0.95) for T cell-mediated rejection; >0.96 (0.90-1.00) with all three for interstitial fibrosis-tubular atrophy. We also developed a classifier to discriminate active and chronic active ABMR with 95% accuracy. In conclusion, we built highly sensitive and specific artificial intelligence classifiers able to interpret kidney graft scoring together with a few clinical data and automatically diagnose rejection, with excellent concordance with the Banff rules and reference diagnoses made by a group of experts. Some discrepancies may point toward possible improvements that could be made to the Banff classification.

Optimization of Vancomycin Initial Dose in Term and Preterm Neonates by Machine Learning.

INTRODUCTION: Vancomycin is one of the antibiotics most used in neonates. Continuous infusion has many advantages over intermittent infusions, but no consensus has been achieved regarding the optimal initial dose. The objectives of this study were: to develop a Machine learning (ML) algorithm based on pharmacokinetic profiles obtained by Monte Carlo simulations using a population pharmacokinetic model (POPPK) from the literature, in order to derive the best vancomycin initial dose in preterm and term neonates, and to compare ML performances with those of an literature equation (LE) derived from a POPPK previously published. MATERIALS AND METHODS: The parameters of a previously published POPPK model of vancomycin in children and neonates were used in the mrgsolve R package to simulate 1900 PK profiles. ML algorithms were developed from these simulations using Xgboost, GLMNET and MARS in parallel, benchmarked and used to calculate the ML first dose. Performances were evaluated in a second simulation set and in an external set of 82 real patients and compared to those of a LE. RESULTS: The Xgboost algorithm yielded numerically best performances and target attainment rates: 46.9% in the second simulation set of 400-600 AUC/MIC ratio vs. 41.4% for the LE model (p = 0.0018); and 35.3% vs. 28% in real patients (p = 0.401), respectively). The Xgboost model resulted in less AUC/MIC > 600, thus decreasing the risk of nephrotoxicity. CONCLUSION: The Xgboost algorithm developed to estimate the initial dose of vancomycin in term or preterm infants has better performances than a previous validated LE and should be evaluated prospectively.

Toxoplasmosis in patients with an autoimmune disease and immunosuppressive agents: A multicenter study and literature review.

BACKGROUND: Cases of Toxoplasma reactivation or more severe primary infection have been reported in patients receiving immunosuppressive (IS) treatment for autoimmune diseases (AID). The purpose of this study was to describe features of toxoplasmosis occurring in patients with AID treated by IS therapy, excluded HIV-positive and transplant patients. METHODS: A multicenter descriptive study was conducted using data from the French National Reference Center for Toxoplasmosis (NRCT) that received DNA extracts or strains isolated from patients, associated with clinical data. Other cases were retrieved through a questionnaire sent to all French parasitology and internal medicine departments. Furthermore, a systematic literature review was conducted. RESULTS: 61 cases were collected: 25 retrieved by the NRCT and by a call for observations and 36 from a literature review. Half of the cases were attributed to reactivation (50.9%), and most of cases (49.2%) were cerebral toxoplasmosis. The most common associated AID were rheumatoid arthritis (28%) and most frequent treatments were antimetabolites (44.3%). Corticosteroids were involved in 60.7% of cases. Patients had a favorable outcome (50.8%) but nine did not survive. For 12 cases, a successful Toxoplasma strain characterization suggested the possible role of this parasitic factor in ocular cases. CONCLUSION: Although this remains a rare condition, clinicians should be aware for the management of patients and for the choice of IS treatment.

Renin-Angiotensin-System Inhibitors for the Prevention of Chemotherapy-Induced Peripheral Neuropathy: OncoToxSRA, a Preliminary Cohort Study.

Chemotherapy-induced peripheral neuropathy (CIPN) is a frequent and dose-limiting adverse side effect of treatment. CIPN affects the oncological prognosis of patients, as well as their quality of life. To date, no specific pharmacological therapy has demonstrated effectiveness in preventing CIPN. Accumulating preclinical evidence suggests that renin-angiotensin system (RAS) inhibitors may have neuroprotective effects. One hundred and twenty patients were included in this observational study and were followed from the beginning of their neurotoxic chemotherapy schedule until their final assessment, at least one month after its cessation. The National Cancer Institute's common toxicity criteria 4.0 (NCI-CTC 4.0) were used to grade the severity of adverse events. Follow-ups also included electrochemical skin conductance and scales for pain, quality of life and disability. Among patients receiving a platinum-based regimen, the mean grade of sensory neuropathy (NCI-CTC 4.0) was significantly lower in the RAS inhibitor group after the end of their anticancer treatment schedule. Because of the observational design of the study, patients in the RAS inhibitor group cumulated comorbidities at risk of developing CIPN. Randomized controlled trials in platinum-based regimens would be worth conducting in the future to confirm the neuroprotective potential of RAS inhibitors during chemotherapy.

Machine learning algorithms to estimate everolimus exposure trained on simulated and patient pharmacokinetic profiles.

Everolimus is an immunosuppressant with a small therapeutic index and large between-patient variability. The area under the concentration versus time curve (AUC) is the best marker of exposure but measuring it requires collecting many blood samples. The objective of this study was to train machine learning (ML) algorithms using pharmacokinetic (PK) profiles from kidney transplant recipients, simulated profiles, or both types, and compare their performance for everolimus AUC0-12h estimation using a limited number of predictors, as compared to an independent set of full PK profiles from patients, as well as to the corresponding maximum a posteriori Bayesian estimates (MAP-BE). XGBoost was first trained on 508 patient interdose AUCs estimated using MAP-BE, and then on 500-10,000 rich interdose PK profiles simulated using previously published population PK parameters. The predictors used were: predose, ~1 h, and ~2 h whole blood concentrations, differences between these concentrations, relative deviations from theoretical sampling times, morning dose, patient age, and time elapsed since transplantation. The best results were obtained with XGBoost trained on 5016 simulated profiles. AUC estimation achieved in an external dataset of 114 full-PK profiles was excellent (root mean squared error [RMSE] = 10.8 µg*h/L) and slightly better than MAP-BE (RMSE = 11.9 µg*h/L). Using more profiles (n = 10,035) did not improve the ML algorithm performance. The contribution of mixing patient and simulated profiles was significant only when they were in balanced numbers, with ~500 for each (RMSE = 12.5 µg*h/L), compared with patient data alone (RMSE = 18.0 µg*h/L).

A Machine Learning Approach to Predict Interdose Vancomycin Exposure.

INTRODUCTION: Estimation of vancomycin area under the curve (AUC) is challenging in the case of discontinuous administration. Machine learning approaches are increasingly used and can be an alternative to population pharmacokinetic (POPPK) approaches for AUC estimation. The objectives were to train XGBoost algorithms based on simulations performed in a previous POPPK study to predict vancomycin AUC from early concentrations and a few features (i.e. patient information) and to evaluate them in a real-life external dataset in comparison to POPPK. PATIENTS AND METHODS: Six thousand simulations performed from 6 different POPPK models were split into training and test sets. XGBoost algorithms were trained to predict trapezoidal rule AUC a priori or based on 2, 4 or 6 samples and were evaluated by resampling in the training set and validated in the test set. Finally, the 2-sample algorithm was externally evaluated on 28 real patients and compared to a state-of-the-art POPPK model-based averaging approach. RESULTS: The trained algorithms showed excellent performances in the test set with relative mean prediction error (MPE)/ imprecision (RMSE) of the reference AUC = 3.3/18.9, 2.8/17.4, 1.3/13.7% for the 2, 4 and 6 samples algorithms respectively. Validation in real patient showed flexibility in sampling time post-treatment initiation and excellent performances MPE/RMSE<1.5/12% for the 2 samples algorithm in comparison to different POPPK approaches. CONCLUSIONS: The Xgboost algorithm trained from simulation and evaluated in real patients allow accurate and precise prediction of vancomycin AUC. It can be used in combination with POPPK models to increase the confidence in AUC estimation.

Estimation of drug exposure by machine learning based on simulations from published pharmacokinetic models: The example of tacrolimus.

We previously demonstrated that Machine learning (ML) algorithms can accurately estimate drug area under the curve (AUC) of tacrolimus or mycophenolate mofetil (MMF) based on limited information, as well as or even better than maximum a posteriori Bayesian estimation (MAP-BE). However, the major limitation in the development of such ML algorithms is the limited availability of large databases of concentration vs. time profiles for such drugs. The objectives of this study were: (i) to develop a Xgboost model to estimate tacrolimus inter-dose AUC based on concentration-time profiles obtained from a literature population pharmacokinetic (POPPK) model using Monte Carlo simulation; and (ii) to compare its performance with that of MAP-BE in external datasets of rich concentration-time profiles. The population parameters of a previously published PK model were used in the mrgsolve R package to simulate 9000 rich interdose tacrolimus profiles (one concentration simulated every 30 min) at steady-state. Data splitting was performed to obtain a training set (75%) and a test set (25%). Xgboost algorithms able to estimate tacrolimus AUC based on 2 or 3 concentrations were developed in the training set and the model with the lowest RMSE in a ten-fold cross-validation experiment was evaluated in the test set, as well as in 4 independent, rich PK datasets from transplant patients. ML algorithms based on 2 or 3 concentrations and a few covariates yielded excellent AUC estimation in the external validation datasets (relative bias < 5% and relative RMSE < 10%), comparable to those obtained with MAP-BE. In conclusion, Xgboost machine learning models trained on concentration-time profiles simulated using literature POPPK models allow accurate tacrolimus AUC estimation based on sparse concentration data. This study paves the way to the development of artificial intelligence at the service of precision therapeutic drug monitoring in different therapeutic areas.

Mycophenolic Acid Exposure Prediction Using Machine Learning.

Therapeutic drug monitoring of mycophenolic acid (MPA) based on area under the curve (AUC) is well-established and machine learning (ML) approaches could help to estimate AUC. The aim of this work is to estimate the AUC of MPA in organ transplant patients using extreme gradient boosting (Xgboost R package) ML models. A total of 12,877 MPA AUC from 0 to 12 hours (AUC0-12 h ) requests from 6,884 patients sent to our Immunosuppressant Bayesian Dose Adjustment expert system (https://abis.chu-limoges.fr) for AUC estimation and dose recommendation based on MPA concentrations measured at least at three sampling times (~ 20 minutes, 1 and 3 hours after dosing) were used to develop two ML models based on two or three concentrations. Data were split into a training set (75%) and a test set (25%) and the Xgboost models in the training set with the lowest root mean squared error (RMSE) in a 10-fold cross-validation experiment were evaluated in the test set and in 4 independent full-pharmacokinetic (PK) datasets from renal or heart transplant recipients. ML models based on two or three concentrations, differences between these concentrations, relative deviations from theoretical times of sampling, presence of a delayed absorption peak, and five covariates (dose, type of transplantation, associated immunosuppressant, age, and time between transplantation and sampling) yielded accurate AUC estimation performances in the test datasets (relative bias < 5% and relative RMSE < 20%) and better performance than MAP Bayesian estimation in the four independent full-PK datasets. The Xgboost ML models described allow accurate estimation of MPA AUC0-12 h and can be used for routine exposure estimation and dose adjustment and will soon be implemented in a dedicated web interface.